Research Methodology in Social Sciences
Dr. David Sichinava
November 16, 2018
Dependent and independent samples
- Two samples are dependent if natural matching occurs between two groups. For instance, repeated measurements of same subjects
- Samples are independent if they are drawn independently.
Compare means of two independent groups: confidence intervals
- For large samples, or small samples from random distributions the sampling distribution of \( (\bar{y}_{2}-\bar{y}_{1}) \) has a normal shape, we use t-distribution to estimate standard errors and confidence intervals
- Confidence interval for \( \mu_{2}-\mu_{1} \) will be:
- \( (\bar{y}_{2}-\bar{y}_{1})+/- t(se) \), where
\( se=\sqrt{\frac{{s_{1}^2}}{{n}_{1}}+\frac{{s_{2}^2}}{{n}_{2}}} \)
- Usually, interpreting differences between two groups is usually safer and more interpretable than testing null hypothesis
Compare means of two independent groups: null hypothesis testing
- \( t=\frac{(\bar{y}_{2}-\bar{y}_{1})-0}{se} \)
- Here we compute test statistic and then look for corresponding p-value
Compare two dependent proportions
- Fisher's exact test OR Chi-square test
What about different scales?
What about different scales?
What happens if the assumptions (of normality) are not met?
- We use non-parametric tests, such as Wilcoxon-Mann-Whitney test
- Instead of comparing means, we compare median values